Using Machine Learning to Detect Potentially Infectious Coronaviruses to Humans

A team of researchers has developed a machine learning method to detect coronaviruses that have the potential to infect humans. The method assigns a score, between 0 and 1, to spike proteins of alpha and beta coronaviruses based on the likelihood of human receptor binding. Known as h-BiP, the score is calculated using trimer embeddings and a logistic regression classifier. Viruses with an h-BiP score of 0.5 or higher are considered likely to bind a human receptor. The method achieved a 99% accuracy rate in testing and identified three bat coronaviruses with unknown binding status that may have the potential to infect humans. Molecular dynamics simulations later confirmed in silico binding and determined contact residues. The h-BiP method could help in detecting potential human-infecting novel coronaviruses. The study has been published in Scientific Reports.